Ensemble Research Based On Artificial Bee Colony Algorithm And Support Vector Machine

Support vector machine(SVM)is a pattern classification technology based on statistical learning theory and it is suitable for handling with the classification problem with small samples,which has been widely applied in various fields of pattern recognition.The performance of SVM classifier is largely affected by parameters and features used,usually tuning of parameters and feature selection are solved separately,it is hard to obtain the optimal SVM classifier overall.Further,because of the complexity of the practical problem,generalization ability of SVM is also needed to be further improved.The ensemble learning provides a new way to improve the generalization ability of the classification system,through training and combing multiple classifiers with different features,it improves the classification performance of classifiers and has made great progress,meanwhile the related research work is not perfect,which is worthy of further study.Starting from the point mentioned above,the thesis mainly studies dealing with the parameter optimization and feature selection simultaneously for SVM by using artificial colony algorithm.First of all,how to deal with parameter optimization of SVM and feature selection for SVM with artificial bee colony algorithm(ABC)is studied;moreover,parameters optimization and feature selection of SVM is handled synchronously as an optimization problem by ABC.The method could increase the classification accuracy of SVM and select less features as far as possible,moreover,it could obtain the overall performance SVM parameters and the optimal feature subset.In order to further improve the generalization ability of SVM,ensemble learning with weighted voting is utilized after completing optimization of parameters and features of SVM.Aiming at obtaining the better classification performance,it builds a number of SVM classifiers respectively,after each SVM classifier is learned,a number of different SVM classifiers are obtained,and the weight of the single SVM classifier is set according to the ratio of the accuracy of each SVM and the total number of classifiers,several different SVM classifiers are combined by weighted voting rules.In order to verify the proposed method,some UCI data sets are utilized,as well this thesis contrasts ABC algorithm with commonly used genetic algorithm(GA)and particle swarm optimization algorithm(PSO).The experimental results show that compared with GA algorithm and PSO algorithm,the ABC algorithm has the better performance for optimization of SVM classifier;further,SVM ensemble based on ABC and weighted voting has good adaptability and classification accuracy,which is able to improve the performance of the basic SVM classifier and select fewer number of features,get the overall performance of the optimal SVM parameters and feature subset.